idea-reality-mcp

English | 繁體中文

idea-reality-mcp

Your AI agent checks before it builds. Automatically.

The only MCP tool that searches 5 real databases before your agent writes a single line of code. No manual search. No forgotten step. Just facts.

Works with: Claude Desktop · Claude Code · Cursor · Windsurf · any MCP client

Try it in your browser — no install

What it does

You: "AI code review tool"

idea-reality-mcp:
├── reality_signal: 92/100
├── trend: accelerating ↗
├── market_momentum: 73/100
├── GitHub repos: 847 (45% created in last 6 months)
├── Top competitor: reviewdog (9,094 ⭐)
├── npm packages: 56
├── HN discussions: 254 (trending up)
└── Verdict: HIGH — market is accelerating, find a niche fast

One score. Five sources. Trend detection. Your agent decides what to do next.

The problem

Every developer has wasted days building something that already exists with 5,000 stars on GitHub.

You ask ChatGPT: "Is there already a tool that does X?"

ChatGPT says: "That's a great idea! There are some similar tools, but you can definitely build something better!"

That's not validation. That's cheerleading.

"Why not just Google it?"

This is the most common question we get. Here's the honest answer:

Google works — if you remember to use it. The problem isn't search quality. The problem is that your AI agent never Googles anything before it starts building.

idea-reality-mcp runs inside your agent. It triggers automatically. The search happens whether you remember or not.

	Google	ChatGPT / SaaS validators	idea-reality-mcp
Who runs it	You, manually	You, manually	Your agent, automatically
Input	You craft the query	Natural language	Natural language
Output	10 blue links — you interpret	"Sounds promising!"	Score 0-100 + evidence + competitors
Sources	Web pages	None (LLM generation)	GitHub + HN + npm + PyPI + PH
Cross-platform	Search each site separately	N/A	5 sources in parallel, one call
Workflow	Copy-paste between tabs	Separate app	MCP / CLI / API / CI
Verifiable	Yes (manual)	No	Yes (every number has a source)
Price	Free	Free trial → paywall	Free & open-source (MIT)

TL;DR — You don't use it. Your agent does. That's the point.

Try it (30 seconds)

uvx idea-reality-mcp

Or try it in your browser — no install, instant results.

Install

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "idea-reality": {
      "command": "uvx",
      "args": ["idea-reality-mcp"]
    }
  }
}

Config file location

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop. You'll see idea_check in the 🔨 tools menu. Try asking:

• "Check if someone has already built a fitness tracking MCP server"
• "Is there competition for an AI-powered invoice parser?"
• "Before I start, run a reality check on: open-source Slack alternative for small teams"

Claude Code

claude mcp add idea-reality -- uvx idea-reality-mcp

Then ask Claude:

• "Check if this idea already exists: CLI tool that converts Figma to React"
• "Run a deep reality check on AI-powered code review tools"

Cursor / Other MCP Clients

Add to .cursor/mcp.json (or your client's MCP config):

{
  "mcpServers": {
    "idea-reality": {
      "command": "uvx",
      "args": ["idea-reality-mcp"]
    }
  }
}

Smithery (Remote)

npx -y @smithery/cli install idea-reality-mcp --client claude

Optional: Environment variables

export GITHUB_TOKEN=ghp_...        # Higher GitHub API rate limits
export PRODUCTHUNT_TOKEN=your_...  # Enable Product Hunt (deep mode)

Optional: Agent auto-trigger

The MCP tool description already tells your agent what idea_check does. To make it run proactively (before every new project), add one line to your CLAUDE.md, .cursorrules, or .github/copilot-instructions.md:

When starting a new project, use the idea_check MCP tool to check if similar projects already exist.

See templates/ for all platforms.

Usage

"I have a side project idea — should I build it?"

Tell your AI agent:

Before I start building, check if this already exists:
a CLI tool that converts Figma designs to React components

The agent calls idea_check and returns: reality_signal, top competitors, and pivot suggestions.

"Find competitors and alternatives"

idea_check("open source feature flag service", depth="deep")

Deep mode scans all 5 sources in parallel — GitHub repos, HN discussions, npm packages, PyPI packages, and Product Hunt — and returns ranked results.

"Build-or-buy sanity check before a sprint"

We're about to spend 2 weeks building an internal error tracking tool.
Run a reality check first.

If the signal comes back at 85+ with mature open-source alternatives, you just saved your team 2 weeks.

New: AI-powered search intelligence

Claude Haiku 4.5 generates optimal search queries from your idea description — in any language — with automatic fallback to our dictionary pipeline.

	Before	Now
English ideas	✅ Good	✅ Good
Chinese / non-English ideas	⚠️ Dictionary lookup (150+ terms)	✅ Native understanding
Ambiguous descriptions	⚠️ Keyword matching	✅ Semantic extraction
Reliability	100% (no external API)	100% (graceful fallback to dictionary)

The LLM understands your idea. The dictionary is your safety net. You always get results.

Tool schema

idea_check

Parameter	Type	Required	Description
idea_text	string	yes	Natural-language description of idea
depth	"quick" | "deep"	no	"quick" = GitHub + HN (default). "deep" = all 5 sources in parallel

Output: reality_signal (0-100), trend (accelerating/stable/declining), sub_scores{} (incl. market_momentum), duplicate_likelihood, evidence[], top_similars[], pivot_hints[], meta{}

Full output example

{
  "reality_signal": 72,
  "duplicate_likelihood": "high",
  "evidence": [
    {"source": "github", "type": "repo_count", "query": "...", "count": 342},
    {"source": "github", "type": "max_stars", "query": "...", "count": 15000},
    {"source": "hackernews", "type": "mention_count", "query": "...", "count": 18},
    {"source": "npm", "type": "package_count", "query": "...", "count": 56},
    {"source": "pypi", "type": "package_count", "query": "...", "count": 23},
    {"source": "producthunt", "type": "product_count", "query": "...", "count": 8}
  ],
  "top_similars": [
    {"name": "user/repo", "url": "https://github.com/...", "stars": 15000, "description": "..."}
  ],
  "pivot_hints": [
    "High competition. Consider a niche differentiator...",
    "The leading project may have gaps in...",
    "Consider building an integration or plugin..."
  ],
  "meta": {
    "sources_used": ["github", "hackernews", "npm", "pypi", "producthunt"],
    "keyword_source": "llm",
    "depth": "deep",
    "version": "0.5.0"
  }
}

Scoring weights

Mode	GitHub repos	GitHub stars	HN	npm	PyPI	Product Hunt
Quick	60%	20%	20%	—	—	—
Deep	25%	10%	15%	20%	15%	15%

If Product Hunt is unavailable (no token), its weight is redistributed automatically.

REST API

Not using MCP? Call the hosted API directly:

curl -X POST https://idea-reality-mcp.onrender.com/api/check \
  -H "Content-Type: application/json" \
  -d '{"idea_text": "AI code review tool", "depth": "quick"}'

Returns the same reality_signal, evidence, and competitors as the MCP tool. Free, no API key required.

CI: Auto-check on Pull Requests

Use idea-check-action to validate new feature proposals:

name: Idea Reality Check
on:
  issues:
    types: [opened]

jobs:
  check:
    if: contains(github.event.issue.labels.*.name, 'proposal')
    runs-on: ubuntu-latest
    steps:
      - uses: mnemox-ai/idea-check-action@v1
        with:
          idea: ${{ github.event.issue.title }}
          github-token: ${{ secrets.GITHUB_TOKEN }}

Roadmap

• v0.1 — GitHub + HN search, basic scoring
• v0.2 — Deep mode (npm, PyPI, Product Hunt), improved keyword extraction
• v0.3 — 3-stage keyword pipeline, 150+ Chinese term mappings, synonym expansion, LLM-powered search (Render API)
• v0.4 — Email gate, Score History, Agent Templates, GitHub Action
• v0.5 — Temporal signals (trend detection and timing analysis)
• v1.0 — Idea Memory Dataset (opt-in anonymous logging)

Found a blind spot?

If the tool missed obvious competitors or returned irrelevant results:

• Open an issue with your idea text and the output
• We'll improve the keyword extraction for your domain

FAQ

How is this different from just Googling? Google requires you to manually search. idea-reality-mcp runs automatically inside your AI agent — no human intent needed. It searches 5 structured databases, not web pages, and returns a scored signal instead of links.

What databases does it scan? GitHub repositories, Hacker News posts, npm packages, PyPI packages, and Product Hunt launches. Quick mode scans GitHub + HN. Deep mode scans all five.

Is it free? Yes. MIT license, open source. The MCP server runs locally. The web demo at mnemox.ai/check is also free.

Does it work for non-English ideas? Yes. The keyword extraction supports Chinese (150+ term mappings) and works with any language input. The Render API uses LLM extraction for better multilingual support.

How does the 0-100 scoring work? The reality signal combines weighted scores from each source — repository count, star count, discussion volume, package downloads. Higher means more existing competition. The formula is intentionally simple and explainable, not ML-based.

License

MIT — see LICENSE

Contact

Built by Mnemox AI · [e-mail]